Analysis of the Recursive Locally-Adaptive Filtration of 3D Tensor Images

This work is focused on the computational complexity (CC) reduction of locally-adaptive processing 3D tensor images, based recursive approaches. As a basis, local averaging operation used, implemented as sliding mean filter (SM3DF) with central symmetric kernel. Symmetry plays very important role in constructing working window. The presented theoretical approach could be adopted various algorithms for processing, such additive noise reduction, sharpness enhancement, texture segmentation, etc. basic characteristics SM3DF are analyzed, together main features adaptive filtration Gaussian noises and unsharp masking where aimed at. In paper, ability implementation through 1D filters, sequentially bonded together, also introduced. evaluated non-recursive mode. suits convolutional neural networks which comprise their layers. result lower CC, promising opportunity opened higher efficiency image networks.